Address correspondence to Vadym Gnatkovsky, Unit of Experimental Neurophysiology and Epileptology, Fondazione Istituto Neurologico, Milano, Italy. E-mail email@example.com
Purpose: The identification of the epileptogenic zone (EZ) is crucial for planning epilepsy surgery in patients with drug-resistant partial epilepsy. This task may require intracerebral encephalography (EEG) monitoring, the results of which are usually interpreted by visual presurgical inspection. A computer-assisted method for rapidly identifying reproducible ictal patterns based on the analysis of time, frequency, and spatial domains of stereo-EEG (SEEG) signals is described here.
Methods: A new method for EZ detection was tested on SEEG recordings performed by intracerebral electrodes in eight patients with pharmacoresistant partial epilepsy. SEEG data were exported to a program developed in LabView.
Key Findings: Prevalent frequencies during seizure events were evaluated by Fourier transform and further integral algorithms. Different frequencies and the relative powers were simultaneously evaluated in all recording leads. Patterns characterized by specific and prevalent frequencies were identified in a subset of recording sites during both seizure onset and seizure development. Three-dimensional (3D) maps of the measurements obtained from each recording channel were reconstructed on magnetic resonance coordinates to visualize the spatial distribution of the EZ. With this method, the reproducibility of ictal patterns in the same patient was characterized. The boundaries of the EZ identified with this algorithm correlated well with the EZ recognized with the traditional approach (n = 8). The spatial distribution of specific SEEG signals associated with different types of seizures was also analyzed in two patients.
Significance: We describe a computer-assisted method to acquire information on EZ boundaries and to verify reproducibility of seizure patterns from intracerebral recordings performed in patients with pharmacoresistant partial epilepsies.
Patients diagnosed with partial pharmacoresistant epilepsy are potential candidates for surgical removal of the epileptogenic zone (EZ), which is defined in clinical practice by employing both electroclinical and imaging criteria. In one third of drug-resistant patients, the cerebral areas responsible for seizure generation can be defined exclusively by intracranial recording with intracerebral or subdural electrodes (Engel, 1990). The correct identification of the EZ on the intracranial electroencephalography (EEG) has a direct impact on postsurgical outcome. Therefore, an accurate analysis of ictal and interictal intracranial EEG during and between seizures is mandatory before surgery.
Following the theoretical principles of stereo-EEG (SEEG) methodology (Munari & Bancaud, 1987), the EZ is considered here as the “site of the beginning and of the primary organization of the epileptic seizures.” Therefore, we focus mainly on the identification of the brain regions in which specific ictal patterns develop during seizures (Bancaud et al., 1970; Engel, 1994). At present, the identification of the EZ is based on the ability of experienced clinical neurophysiologists to identify pathologic intracranial EEG patterns by visual inspection. This process is time-consuming and requires the full-time involvement of extremely gifted and dedicated neurophysiologists. A quantified analysis of the intracranial EEG may help to accelerate the process of EZ identification and should reduce the bias due to experience of the operator responsible for the EEG report. Computer-assisted analysis of intracranial signals may also help to resolve the issue of complexity associated with the massive amount of data recorded during presurgical intracranial EEG sessions. Finally, a quantified evaluation of intracranial EEG traces should contribute to understand the neurobiologic mechanisms responsible for the initiation, the progression, and the diffusion of focal seizures in humans.
We describe here a method to rapidly identify reproducible ictal patterns on intracranial EEG signals. We specifically aimed at developing a tool that could be helpful to verify the reproducibility of seizure patterns and to delineate the margins of the EZ in consecutive seizures characterized by similar clinical features recorded in the same patient. We also aimed at correlating different patterns and locations/extensions of the EZ with different types of seizures.
Patient stereo-EEG recordings
SEEG data recorded during presurgical evaluation in eight patients with drug-resistant focal epilepsy were utilized (see patient’s data in Table S1). SEEG recordings were performed using intracranial multichannel electrodes (DIXI Medical, 5–18 contacts; length, 2 mm, diameter, 0.8 mm; 1.5 mm apart) intracerebrally implanted according to the methodology introduced by the group of the Saint Anne Hospital in Paris (France) based on the stereotactic method of Talairach (Talairach et al., 1974; Cossu et al., 2005). The position of the implanted electrodes was determined in each patient from the available clinical and instrumental noninvasive information. Signals sampled at 1 kHz were recorded on a Neurofax EEG-1100 system (Nihon Koden, Tokyo, Japan) with 5–14 contacts per intracranial electrode and a maximum number of 170 recording channels. Data were digitized to hard disk with 16-bit resolution. SEEG was recorded for 5–10 days under video monitoring to verify the electroclinical features of the seizures. SEEG sample traces utilized in the present report were recorded in patients with focal cortical dysplasia located in extratemporal regions.
Correlation between SEEG and MRI
The actual position of the implanted electrodes was postoperatively determined by computed tomography (CT) scan without contrast. Based on the preoperative magnetic resonance imaging (MRI) data and postimplantation CT, image fusion was performed to locate each lead anatomically along each electrode trajectory.
Monopolar SEEG data were exported in ASCII format and converted to the format of our new software “Elpho-SEEG” (Fig. 1A), developed in LabView (National Instruments, Austin, TX, U.S.A.). To evaluate frequency changes in time, fast Fourier transform (FFT) was calculated for each recorded channel on one second sliding windows; consecutive windows overlapped by 0.2 s (Fig. 1B, Video S1). This approach increases the accuracy of the calculation, even though it introduces a shift between the integral traces and the EEG traces that is proportional to the selected time window duration and overlap. The duration of the sliding window could be modified at will. For FFT computation of very slow frequencies the sliding window was expanded to 3 s. For all recording sites, data were saved as a full frequency range of FFT power spectra for each position of the sliding window. To identify particular frequencies that characterize ictal pattern profiles, these data were scanned as described in the Results. The integral of the power spectrum was calculated for all frequency ranges included between 1 and 250 Hz. Sampled frequencies of interest (FOIs) between 110 and 125 Hz are illustrated in the example in Fig. 1C. Changes of FOI integral values with time were represented either as traces or as color-intensity maps, as show in Fig. 1C). A slight delay is introduced between the FOI integral increase phase illustrated as the integral trace and color-coded strip representation, because of the resolution of the color gradient map. Online (on the fly) evaluation of the changes in frequency content displayed in all recording electrodes was accomplished with the help of a moving slider (Video S2). The program rapidly calculated the set frequencies and updated the color intensity map according to the new calculation (see Results; Figs. 2, 3, and 4).
3D Representation of the EZ
Based on the spatial coordinates of the contacts calculated from MR sequences, a three-dimensional (3D) map of the implanted electrodes was reconstructed. The amplitude of the value calculated by frequency analysis proportionally correlated with the size of the dot that represents each contact (see Figs. 5 and 6, Video S3). By this representation, the EZ area associated with characteristic seizure pattern generation was clearly visualized.
The EZ was identified through a computational approach performed on all seizures recorded in the same patient. Our first aim was to verify the reproducibility of both SEEG patterns and EZ extension during seizures that showed identical clinical features recorded in the same patient. Seizures were identified by video monitoring and SEEG signals included between 20 s before seizure onset and 20 s after seizure termination were selected. Complete seizure profiles were reconstructed from the imported SEEG data (Fig. 1A; see also Fig. 2 and upper panels in Fig. 3).
Seizure patterns and the EZ
Following the power spectrum quantification of frequencies between 1 and 250 Hz in all channels, the typical focal pattern of the ictal period was identified. As shown in Fig. 2, the scan of different frequencies in the SEEG during a seizure revealed a focal pattern characterized by a peak at 8–13 Hz at ictal onset, followed within 3 s by a very fast activity (220–270 Hz) that correlated with a very slow (0.3–3 Hz) component during the EEG phase of voltage decrement. We will refer to each of these identified frequencies as frequency of interest (FOI). FOIs were identified for each patient/seizure after the evaluation of frequency changes occurring in all channels. The recording sites that generated the specific FOI observed at seizure onset (for instance 8–13 Hz in the lower left panel) and the following FOI patterns (0.3–3 and 220–270 Hz in the right panels) defined the extension of the EZ. FOIs that characterized a seizure had to be individually identified for each patient. The seizure illustrated in the example in Fig. 2 terminated with a diffuse high-frequency repetitive discharge (lower right panel, asterisks).
Identification of the EZ
Once the pattern of FOIs associated with a seizure type in one patient was identified, all channels were sorted based on the integral value of the FOI calculated during a defined phase of the seizure (Fig. 3; please note that profiles of integral FOI values are plotted here). In the example illustrated in Fig. 3A, the period included between the vertical bars (marked by the asterisk) was utilized to extract the average integral values that are illustrated for each channel in the right panel in A. The duration of the sliding window could be modified according to the needs. Next, integral values were sorted on the basis of their amplitude (higher values on bottom), as illustrated in Fig. 3B. In Fig. 3B, the single FOI integral traces are also sorted accordingly. The observation of a pool of 5–10 contacts with significantly higher integral values suggests that the specific FOI (100–125 Hz in this patient) has a focal origin. FOI integral values were evaluated by analyzing the normal (Gaussian) distribution range. We selected as potential generators of the FOI those contacts that showed values higher than the normal Gaussian distribution demonstrated in the rest of the contacts (see Fig. 5 for details). We assumed that this group of contacts represents the focal region that generates the observed pattern.
Interseizure stability of EZ extension in the same patient
Next, we evaluated whether seizures that showed identical clinical features correlated with similar spatial distribution of FOIs and, therefore, were generated within the same EZ. A similar, very focal FOI pattern at 10–20 Hz was observed in all eight seizures recorded in the patient shown in Fig. 4A. The same contacts (n. 110–114, 82 on electrode J' and n. 10 on electrode N' in Fig. 4B) were involved in FOI generation in all eight seizures. To illustrate the reproducibility of FOI distribution in consecutive seizures, a dot was positioned on the representation of the recording contacts illustrated in Fig. 4B, when significant (see Fig. 5A); log-normal distribution of FOI integral values was then evaluated. The result of this procedure suggests the existence of a highly reproducible and focal pattern for the FOI identified as typical of seizure onset in this patient.
Next, we plotted the distribution of the mean FOI integral values on a schematic representation of the implanted SEEG electrodes. Panel A in Fig. 5 illustrates the mean integral values of FOI at 10–20 Hz for all recorded contacts measured in the eight seizures shown in Fig. 4. The analysis of the log-normal distribution of this parameter (Fig. 5A, bottom graph) suggests that the FOI mean power integral of the seven contacts on the right of the arrow (>95% area of the distribution) has significantly (p > 0.05) higher values than the rest of the contacts. The values of these seven suprathreshold contacts are better illustrated in the right panel in Fig. 5A. As described in the Methods, the relative position of the intracranially implanted SEEG electrodes was reconstructed following computed tomography-magnetic resonance imaging (CT-MRI) data fusion. The 3D reconstruction of the SEEG electrode position is illustrated in the Video S3. Figure 5B, shows the data shown in 5A superimposed on the implanted electrodes scheme and a 3D Cartesian space that reproduced the position of the SEEG electrodes. The size of the dots that represent the recording sites correlates with the mean integral FOI value. For the analyzed patient, the largest mean integral FOI values were on seven recording sites that belong to three electrodes—J', N', and N, as illustrated in Fig. 5B and in the Video S3. These findings demonstrate that the reproducibility of EZ extension in recurring focal seizures recorded in the same patient can be evaluated with this method.
The FOIs observed at seizure onset (square) and during the following phases of seizures (circles) in the eight patients analyzed with the present computer-assisted method are illustrated in Fig. 6. The extension of the EZ observed in all eight patients during this retrospective analysis correlated well with the EZ identified with the traditional visual inspection of the SEEG recordings. The demonstration that FOIs were specific for seizures recorded in the same patients but were different between patients confirms the complexity of the evaluation of seizure patterns in focal drug-resistant epilepsies. This suggests that seizure patterns can be individually characterized by a quantified approach.
Identification of different EZ/seizures in the same patient
We further tested whether it is feasible to identify different seizures types (and ultimately different EZs) based on the described quantifying approach. In Fig. 6, two types of SEEG seizure profiles recorded in the same patient are shown: seizure type I was recorded five times and seizure type II 13 times during the 1-week period of presurgical SEEG recording. The demonstration of activity around 8–13 Hz in the same contacts during both seizure types (second pair of panels from the top in Fig. 7A), suggested a common generator for this FOI. The average integral FOI values are quantified in the 3D reconstruction of the electrode positions illustrated in the inserts (see also Fig. 7B). As type I seizure progressed, a very high FOI (220–270 Hz) coupled with a slow (0.3–3 Hz) FOI was observed. Both patterns were generated by the same contacts located in an area close to the contacts involved in 8–13 Hz FOI generation (Fig 7A; compare inserts of the second left panel with the third and fourth panels from top). This region coincided with the lesion (focal dysplasia) illustrated in the left preoperative MR (Pre OP) insert of Fig. 7B. Type II seizures showed a different evolution (both clinically and neurophysiologically), characterized by a FOI at 140–170 Hz generated by the contacts located in mesial area (third panel from the top on the right in Fig. 7A). Based on these data, the mesial part was retrospectively proposed to be included in the EZ of type II seizures, but not in type I seizures. As illustrated in Fig. 7B (Post OP insert), the mesial region was interpreted as a secondarily activated generator during the visual identification of the EZ and was excluded from surgery. Distinct seizure patterns with different EZ extensions were analyzed and identified in two patients of our series.
The diagnostic procedure to identify the EZ in patients with drug-resistant focal epilepsies is essentially based on visual inspection of intracranial EEG signals. This approach is time-consuming even for an experienced neurophysiologist, and it is usually based on the analysis of a selected subset of recording contacts (usually 30–40 of >150). For this reason we have developed a novel method to facilitate the unrestricted screening of the massive amount of data that are recorded from a single patient during presurgical intracranial investigation. This method does not substitute for clinical expertise, but contributes to the identification of the EZ. Ideally, quantified analysis of a large quantity of EEG data should grant several potential advantages in terms of precision, objectivity, and reporting time. In addition, computer-assisted EEG analysis might reveal events that are hidden to visual inspection. Finally, the predictable bias due to the experience of the operator should be reduced with quantified analysis; reproducible results should be achieved more easily and by a larger number of neurophysiologists with less extensive and specialized training.
We described here a new method developed in LabView to localize and quantify reproducible/diverse seizure patterns recorded in the same patient with intracranial SEEG electrodes. Pattern identification and EZ localization were based on the study of energy content of all recorded frequencies measured in a large number of recording sites (up to 170). After selection of the intracranial EEG signals corresponding to each seizure event, this procedure helps to delineate the EZ in a relatively short time. In addition, this method facilitates the identification of similarities and differences between seizures observed in the same patient. This is a particularly interesting feature, since the epileptogenic network involved in seizure onset and progression could be distinct in different types of seizures. When more than one seizure patterns is observed in the same patient, several possible mechanisms could be considered. Different epileptogenic networks may be involved at seizure onset in the case of multifocal partial epilepsy. If one epileptogenic network is active, larger/smaller extents of the same network can be recruited. This could be the consequence of several factors that include changes in therapy or in drug dose, changes in vigilance status and diet, and other endogenous and exogenous causes.
EEG analysis has long shown that ictal events correlate with changes in the frequency content of brain activity. For this reason seizure networks were explored by analyzing the existence of periods of switch between dominant frequency patterns that correlated with the onset of the clinical seizure. Identification of the frequencies that characterize different ictal phases and their quantification for each recorded contact was the focus of our study and could be achieved by the application of the method described herein.
The quantified approach to SEEG analysis contributes to define and understand ictogenesis, by extracting spatial and temporal information from the complex signals directly recorded within the epileptic human brain. Other methods have employed signal analysis to automatically detect (Gotman, 1985, 1999) or predict seizures (Mormann et al., 2007). This is not within the scope of the method described here, since our work was aimed at drawing the boundaries of the region involved in the specific signal changes that characterize a seizure.
Time and frequency analysis of SEEG signals were utilized to define the changes in the relationship and interdependencies between EZ subregions during the transition from the interictal to the ictal state (Gotman & Levtova, 1996; Bartolomei et al., 2001, 2008; Wendling et al., 2003). These studies obtained from nonlinear correlation-based or coherence-based methods have shown that preferential interactions occur between subregions of the mesial temporal lobe during the generation of epileptic patterns. In one of these studies (Bartolomei et al., 2008) an epileptogenicity index (EI) was identified on the basis of both spectral and temporal properties of fast activity patterns observed in SEEG signals recorded from patients with mesial temporal lobe epilepsy. Statistically high EI values corresponded with structures involved early in the ictal process and producing rapid discharges at seizure onset.
The evaluation of source dipole mapping of interictal events was proposed in the past for the characterization of the epileptogenic region (Ebersole & Hawes-Ebersole, 2007). The relevance of interictal pattern localization for the definition of the EZ is questioned nowadays, since it has been demonstrated that the brain region that generates interictal potentials is larger than the area that is involved in seizure generation (Alarcon et al., 1997; Hufnagel et al., 2000; de Curtis & Avanzini, 2001).
The present report describes a new computer-assisted method developed to analyze intracranial EEG traces, particularly focused on the identification of reproducible seizure patterns recorded in the same patient. In addition to the obvious clinical and diagnostic implications, the use of quantified analysis of intracranial EEG signals could contribute to understanding of the neurobiologic mechanisms responsible for the initiation and the propagation of focal seizures in humans. The characterization of specific patterns of interictal discharge and seizure initiation/progression/propagation in human focal epilepsy may guide and assist future research on animal models. The reproduction of human seizure patterns in experimental models should help to extract information about the underlying network and cellular mechanisms, thereby improving the translational value of experimental studies on focal ictogenesis.
Acknowledgements and Funding
The study was sponsored by a grant of the Mariani Foundation 2008, and by the Italian Health Ministry Young Investigator Grant (Finanziamento Giovani Ricercatori RF114) to VG. We thank Dr. Suela Dylgjeri and Dr. Chiara Pastori for clinical assessment and discussion during early stage of the study.
We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.
None of the authors has any conflict of interest to disclose.